Unveiling Ancient Maya Settlements Using Aerial LiDAR Image Segmentation

TL;DR

The proposed approach uses novel pre-processing of the raw LiDAR data and dataset augmentation methods to produce trained YOLOv8 networks to improve accuracy, precision, and recall for the segmentation of two important Maya structure types: annular structures and platforms.

Abstract

Manual identification of archaeological features in LiDAR imagery is labor-intensive, costly, and requires archaeological expertise. This paper shows how recent advancements in deep learning (DL) present efficient solutions for accurately segmenting archaeological structures in aerial LiDAR images using the YOLOv8 neural network. The proposed approach uses novel pre-processing of the raw LiDAR data and dataset augmentation methods to produce trained YOLOv8 networks to improve accuracy, precision, and recall for the segmentation of two important Maya structure types: annular structures and platforms. The results show an IoU performance of 0.842 for platforms and 0.809 for annular structures which outperform existing approaches. Further, analysis via domain experts considers the topological consistency of segmented regions and performance vs. area providing important insights. The approach automates time-consuming LiDAR image labeling which significantly accelerates accurate analysis of historical landscapes.

Figures (5)

• Figure 1: (a) Examples of platforms and annular structures visualized in a hillshading image where platforms are outlined in brown and annular structures are in white. (b) Maya structures found in three sites, Muluchtzekel (MLS), Sayil (SAY), and Huntichmul (HNT), in the Puuc region of Yucatan, Mexico were used as data for analysis. (c) The inference result of Huntichmul demonstrates the proposed system's capability of extracting platforms and annular structures. The ground truth and predicted masks were superimposed onto a hillshading image for better visualization. The brown and white polygons respectively outline the ground truth boundaries of platforms and annular structures, while the solid beige and red regions show the predictions of YOLOv8 respectively.
• Figure 2: Pixel count per object for platforms and annular structures from the LiDAR images used in this paper.
• Figure 3: Hillshading image representation of MLS, SAY, and HNT sites where the "bumps" denote hill locations. The images of Muluchtzekel (MLS), Sayil (SAY), and Huntichmul (HNT) have a resolution of 6000$\times$4000, 6000 $\times$8000, and 3600$\times$5000 pixels respectively. The MLS and SAY sites are larger than the HNT despite being visualized in the same size.
• Figure 4: An excerpt of the inference result of HNT where the brown and white polygons respectively indicate the ground truth boundaries of platforms and annular structures and the solid beige and red regions respectively show the predictions of YOLOv8.
• Figure 5: Different ALS data representations.